Digital Twins for Personalized Drug Response Prediction

Abstract

Personalized medicine requires predictive models capable of estimating how individual patients respond to specific medications before treatment begins. Digital twin technology offers a promising solution by constructing virtual patient representations that evolve alongside real physiological changes. However, current digital twin systems often rely on static datasets and limited biological modalities, reducing their clinical applicability. This study proposes a Dynamic MultiModal Digital Twin (DMDT) framework that integrates longitudinal electronic health records, genomic biomarkers, wearable sensor measurements, and laboratory investigations into a continuously updated virtual patient model. The framework combines temporal transformers with graph neural networks to capture both time-dependent physiological variations and complex biological relationships. The proposed methodology aims to improve individualized drug response prediction while supporting adaptive treatment planning. The study addresses existing limitations in multimodal integration and uncertainty-aware prediction, providing a scalable foundation for next-generation AI-assisted precision therapeutics.